PR #30 merged: Engine sync: sfx field + matrix cases, loudness AAC headroom, v5 cut ruleset

Sync from the studio's v5 work (studio PR StuMason/edator-studio#2).

• sfx segment field: timed sound accents — source-time {file, start, gain?}, mixed UNDER speech exactly like bleep tones, projected under speed, truncated at the segment boundary. One ffmpeg input per occurrence.
• validate: file existence (pack-relative), start-in-window, and a new warnings channel (gain > 1 warns).
• test-matrix: 17 cases (adds sfx / sfx+speed / sfx+bleep) — green here.
• loudness: the de-clip pre-AAC target now sits 0.15 dB below the true-peak ceiling; targeting the ceiling exactly let AAC re-expansion ship a -0.99 dBTP master against a -1.0 QC gate.
• SKILL ruleset: comedy air (the exception to cut-tight), device escalation, sound accent economy, the stakes ticker, reaction grammar.

🤖 Generated with Claude Code